Abstract :

The concept of crowd counting is to count the number of people in a locality using a live camera. Here we go with the real time crowd detection of the streaming video, as we can monitor the number of people visiting a particular place for security purpose, surveillance of sports event, political rallies, stampedes and various applications. The factors that reduce the accuracy on crowd counting are similar appearance of people, improper projection of head from the view point of the camera. In this work, switching convolutional neural networks (S-CNN) is used for increasing the accuracy of crowd detection and counting. In earlier approaches the inter scene variation is not considered but while using S-CNN the inter scene variation and semantic analysis is considered to improve the estimation of the count. The Shanghai-Tech data set is used to train and evaluate the CNN model. S-CNN model is an algorithm used to segregate the patches of the frames of the video based on the density of the crowd and to count with more accuracy. The training data set includes 300 images of various places and test images are taken from the frames of the videos.